Claude Opus 5.5's fast mode left research preview on October 15, 2026: Anthropic announced through its official Claude Devs account that the mode has dropped its waitlist and is now generally available to developers on the Claude Platform. It runs the same model with the same weights, just on a faster inference configuration, delivering up to 2.5 times the output speed of standard mode. The price doubles for that speed: 8 dollars per million input tokens and 40 dollars per million output tokens, against 4 and 20 dollars in standard mode.
What gets faster is output, not time to first token
Fast mode raises the number of output tokens produced per second. The official documentation is explicit that the wait for the first token does not shrink in the same way. That makes it a fit for long outputs where someone is waiting on the result: long-running coding agents, tasks that stream out large amounts of analysis, and interactive products built around streaming replies. Developers opt in with a speed setting on the request, and the response reports which speed tier actually served it.
The mode also has its own dedicated rate-limit pool, accounted separately from standard limits. For Opus 5.5, every usage tier carries a default fast mode allowance, so there is no separate access request to file — that is the most concrete change in leaving preview: instead of queueing for a slot, teams can simply turn the parameter on. When the pool is exhausted, the API returns a clear retry time, and capacity typically frees up after a short wait.
Double the price buys waiting time
Whether to switch it on depends on how a workload keeps its accounts. The longer and more sequential the output, the more real the savings: in an agent task with hundreds of steps, a small gain at every step adds up to a whole block of time. Offline batch work, by contrast, should never use fast mode — the two cannot be combined anyway, and batch processing at half price is the better tool there.
There is also a pitfall that is easy to miss: switching the same conversation between fast and standard speeds invalidates the prompt cache. Long-context workloads that rely on caching to control costs can see the time saved eaten by recomputation if they flip tiers too often. Anthropic also notes that fast mode cannot be combined with a Priority Tier commitment, and is not yet available on the AWS-hosted version of the Claude Platform.
Cloud rollout: Microsoft first, Google next
According to the announcement, fast mode is now live on Microsoft's Foundry hosting service, with Google Cloud support coming; fast mode inside Claude Code stays as it was and continues to draw on extra usage credits. For enterprises buying Claude through cloud marketplaces, the significance is less the speed itself than the new shape of the offer: the most expensive flagship model now has an express lane that can be switched on per task — standard pricing for routine work, double pricing only when a deadline justifies it. Anthropic's recent enterprise moves keep circling control and trust; Claude's integration of official voting information is the same logic from another angle: get controllability right first, then scale.
